Look at HTML::Tree. I don't think this gives you an XPATH approach out of the box but it's not far off in it's representation of an HTML page as a tree. You should be able to scan the documents pretty easily.